Quantitative analysis of activation patterns in fMRI images of language regions of interest in the human brain in healthy subjects and patients with epilepsy

This dissertation describes the development of novel methods that use computer aided analysis of functional magnetic resonance imaging (fMRI) data to quantify the language activation patterns in certain regions of interest (ROI) in patients with epilepsy, and distinguish them from normal activation patterns. Previous studies attempted to study activation development by mining simple features and tracking them through development. As a result, none of these studies were able to clearly quantify the numerous differences in brain activation between normal populations and epilepsy patients' populations at different ages. Activation is a complex, multifaceted topic. Consequently, complex non-linear features must be studied and examined in detail. Various analysis techniques were used to investigate the difference between normal and patient activation patterns when performing the same task. Functional MRI images of both normal individuals and patients of different epilepsy types contain a wealth of information that have yet to be explored. By extracting volumetric and topological activation features from normal images and tracking how those features develop through age, the functional development of language networks in the brain can be followed through age especially in the developing ages (4-12 years) of normal individuals, and this can provide information regarding how epilepsy perturbs the development of a normal language network. There were two goals for this project; the first goal of this work was to develop a set of tools to evaluate any set of fMRI data that has been normalized to the MNI brain template. The other major goal was to use these tools to test the synaptic consolidation in normal population through age, and to investigate delays in language development in epilepsy patients compared to the normal population. Achieving these goals was approached by developing new analysis techniques and algorithms capable of identifying a normal pattern of language activation and then comparing epilepsy subjects' trends to that pattern. Two datasets were used in this project, the first was the subjects' data that was thresholded for activation, and the second was the raw (unthresholded) dataset of the same subjects. Nine volumetric features were calculated as well as hundreds of topological features for each region of interest. Correlation Analysis was applied to reduce the high number of features used. One way analysis of variance was then used to investigate differences between different subject groups using features that resulted from correlation analysis. Finally, Euclidean distance comparisons were computed to draw distinction between subject groups in feature space. The work described in this dissertation offers a number of contributions. Two developmental theories were quantitatively inspected for the first time using these calculated features. The first was the synaptic consolidation theory. This hypothesis was tested by calculating differences between features' means of the normal population at different age groups. The outcome of this test was not supportive of the hypothesized statements. Patients' immaturity theory was also examined by computing differences in correlation between normals and patients of matching and of superior age groups. The results of these tests strongly supported this theory. The methods developed in this work provide novel analysis techniques that can be used to investigate complex activation patterns and characteristics. With additional data and further investigation, it will be possible to test neurological development of subjects with a wider age range and more variability to explore normal activation patterns as well as those of patients.